NovelESN

A novel Echo State Network (ESN) algorithm for predicting very complicated time series without
the numerical instability issues of the standard one.

main.py

It receives a txt file with a time series (training data) and it predicts its continuation at certain future instants of time. The program will save and plot the prediction obtained. The code is run via terminal:

python main.py training_file.txt outputfile.txt tau history_q

The user can also specify the optional arguments beta, test_file and history_p, as shown in section examples.

• training_file.txt (string type): Path to a txt file with a single column of values representing a time series of M scalars {y₀, y₁, ...y_M-1}.
• outputfile (string type): Path to the output file where the prediction will be written.
• tau (integer type): Jump in the future. If M-1 is the time index of the last training data value y_M-1, tau indicates that the prediction will end at instant M+tau (last predicted value will be y_M+tau).
• history_q (integer type): Determines the amount of predicted values and the beginning of the prediction. The total number of predicted values will be q and they will go from instant M+tau-(q-1) to M+tau.
• beta (optional, float type, default=1e-10): Regularization parameter for a ridge regression equation. See the file equations_doc.pdf for more info.
• test_file.txt (optional, string type, default=None): Path to a txt file with test data values. The test data is used to know if the prediction is good or not. A perfect prediction should match the test data. If provided, the program will show a plot comparing the prediction and the test data.
• history_p (optional, integer type, default=32): Number of feedback values used to update the reservoir of neurons during training. See equations_doc.pdf for more info.

The user can get a help menu with python main.py -h

To sum up

Given a time series of M scalars {y₀, y₁, ...y_M-1}, the user will specify a jump in the future tau and a history q. The program will then return a set of predicted values {y_M+tau-(q-1), y_M+tau-(q-2), ...y_M+tau}

System Requirements

Python 3.7 or newer with the following libraries:

• numpy
• matplotlib
• sklearn

Examples

The user must respect the order of the positional arguments. However, the optional arguments can be called in any order by using a dash:

python main.py TrainingSignals/dynamo_train.txt dynamo_pred.txt 151 152 -test_file TestSignals/dynamo_test_tau151_q152.txt 

python main.py TrainingSignals/dynamo_train.txt dynamo_pred.txt 80 50 -beta 1e-9

python main.py TrainingSignals/sinus_train.txt sinus_pred.txt 99 50 -test_file TestSignals/sinus_test_tau99_q50.txt 

python main.py TrainingSignals/sinus_train.txt sinus_pred.txt 60 61 -history_p 15